Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
REVIEW                                               CORRECTED 24 APRIL 2009; SEE LAST PAGE                               ...
REVIEWon—as the basic stuff of community, and they nectedness (or density) of the family’s social                 Internat...
REVIEW                                                                                                                    ...
REVIEWcontrast, a node with many structural holes can       value and not compared to expected values gen-            This...
CORRECTIONS & CLARIFICATIONSERRATUM                                                                 Post date 24 April 200...
Upcoming SlideShare
Loading in …5

Network analysis in the social sciences


Published on

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Network analysis in the social sciences

  1. 1. REVIEW CORRECTED 24 APRIL 2009; SEE LAST PAGE terms, making it possible to objectively discover emergent groups in network data (5). Another Network Analysis in the Social Sciences front was the development of a program of lab- oratory experimentation on networks. Researchers at the Group Networks Laboratory at the Massa- Stephen P. Borgatti, Ajay Mehra, Daniel J. Brass, Giuseppe Labianca chusetts Institute of Technology (MIT) began studying the effects of different communication Over the past decade, there has been an explosion of interest in network research across the network structures on the speed and accuracy physical and social sciences. For social scientists, the theory of networks has been a gold mine, with which a group could solve problems (Fig. yielding explanations for social phenomena in a wide variety of disciplines from psychology to 2). The more centralized structures, such as the economics. Here, we review the kinds of things that social scientists have tried to explain using star structure, outperformed decentralized struc- social network analysis and provide a nutshell description of the basic assumptions, goals, and tures, such as the circle, even though it could be explanatory mechanisms prevalent in the field. We hope to contribute to a dialogue among shown mathematically that the circle structure researchers from across the physical and social sciences who share a common interest in had, in principle, the shortest minimum solution understanding the antecedents and consequences of network phenomena. time (6). Why the discrepancy? Achieving the mathematically optimal solution ne of the most potent ideas in the social would have required the nodes to O sciences is the notion that individuals are embedded in thick webs of social rela- tions and interactions. Social network theory LW JN execute a fairly complex sequence of information trades in which no single node served as integrator of provides an answer to a question that has pre- SR HL the information. But the tendency in LS occupied social philosophy since the time of human networks seemed to be for Plato, namely, the problem of social order: how the more peripheral members of a autonomous individuals can combine to create SN network (i.e., the nodes colored blue enduring, functioning societies. Network theory in the “Star,” “Y,” and “Chain” net- also provides explanations for a myriad of social C12 C10 works in Fig. 2) to channel infor- phenomena, from individual creativity to corpo- mation to the most central node (i.e., rate profitability. Network research is “hot” today, the nodes colored red in Fig. 2), with the number of articles in the Web of Science who then decided what the correct on the topic of “social networks” nearly tripling LC HC HIL answer was and sent this answer in the past decade. Readers of Science are already RT back out to the other nodes. The familiar with network research in physics and ZR fastest performing network struc- biology (1), but may be less familiar with what DD tures were those in which the dis- has been done in the social sciences (2). HN HL tance of all nodes from the obvious FL integrator was the shortest (7). History The work done by Bavelas and C3 In the fall of 1932, there was an epidemic of C5 his colleagues at MIT captured the runaways at the Hudson School for Girls in up- Fig. 1. Moreno’s network of runaways. The four largest circles imagination of researchers in a num- state New York. In a period of just 2 weeks, 14 (C12, C10, C5, C3) represent cottages in which the girls lived. ber of fields, including psychology, girls had run away— a rate 30 times higher than Each of the circles within the cottages represents an individual political science, and economics. In the norm. Jacob Moreno, a psychiatrist, suggested girl. The 14 runaways are identified by initials (e.g., SR). All the 1950s, Kochen, a mathematician, the reason for the spate of runaways had less to nondirected lines between a pair of individuals represent feelings and de Sola Pool, a political scien- do with individual factors pertaining to the girls’ of mutual attraction. Directed lines represent one-way feelings tist, wrote a highly circulated paper, personalities and motivations than with the po- of attraction. eventually published in 1978 (8), sitions of the runaways in an underlying social which tackled what is known today network (3). Moreno and his collaborator, Helen of modeling the social sciences after the physical as the “small world” problem: If two persons are Jennings, had mapped the social network at Hudson ones was not, of course, Moreno’s invention. A selected at random from a population, what are using “sociometry,” a technique for eliciting and hundred years before Moreno, the social philos- the chances that they would know each other, graphically representing individuals’ subjective opher Comte hoped to found a new field of and, more generally, how long a chain of acquaint- feelings toward one another (Fig. 1). The links in “social physics.” Fifty years after Comte, the anceship would be required to link them? On the this social network, Moreno argued, provided French sociologist Durkheim had argued that basis of mathematical models, they speculated channels for the flow of social influence and ideas human societies were like biological systems in that in a population like the United States, at least among the girls. In a way that even the girls them- that they were made up of interrelated compo- 50% of pairs could be linked by chains with no selves may not have been conscious of, it was their nents. As such, the reasons for social regularities more than two intermediaries. Twenty years later, location in the social network that determined were to be found not in the intentions of individ- Stanley Milgram tested their propositions empir- whether and when they ran away. uals but in the structure of the social environ- ically, leading to the now popular notion of “six Moreno envisioned sociometry as a kind of ments in which they were embedded (4). Moreno’s degrees of separation” (9). physics, complete with its own “social atoms” sociometry provided a way of making this abstract During this period, network analysis was also and its laws of “social gravitation” (3). The idea social structure tangible. used by sociologists interested in studying the In the 1940s and 1950s, work in social net- changing social fabric of cities. The common con- LINKS Center for Network Research in Business, Gatton College works advanced along several fronts. One front viction at the time was that urbanization destroyed of Business and Economics, University of Kentucky, Lexington, KY 40506–0034, USA. E-mail: (S.P.B.), was the use of matrix algebra and graph theory to community, and that cities played a central role in (A.M.), (D.J.B.), and formalize fundamental social-psychological con- this drama. These sociologists saw concrete rela- (G.L.) cepts such as groups and social circles in network tions between people—love, hate, support, and so892 13 FEBRUARY 2009 VOL 323 SCIENCE
  2. 2. REVIEWon—as the basic stuff of community, and they nectedness (or density) of the family’s social International Network for Social Network Analysis),used network analysis to represent community network. The more connected the network, the an annual conference (Sunbelt), specialized soft-structure. For example, researchers interviewed more likely the couple would maintain a tradi- ware (e.g., UCINET), and its own journal (Social1050 adults living in 50 northern Californian tional segregation of husband and wife roles, Networks). In the 1990s, network analysis radiatedcommunities with varying degrees of urbanism showing that the structure of the larger network into a great number of fields, including physicsabout their social relations (10). The basic pro- can affect relations and behaviors within the dyad. and biology. It also made its way into severalcedure for eliciting network data was to get re- In the 1970s, the center of gravity of network applied fields such as management consultingspondents (egos) to identify people (alters) with research shifted to sociology. Lorrain and White (23), public health (24), and crime/war fightingwhom they had various kinds of relationships (18) sought ways of building reduced models of (25). In management consulting, network analysisand then to also ask ego about the relationships the complex algebras created when all possible is often applied in the context of knowledge manage-between some or all of the alters. They found that compositions of a set of relations were constructed ment, where the objective is to help organizationsurbanism did in fact reduce network density, (e.g., the spouse of the parent of the parent of …). better exploit the knowledge and capabilities dis-which, in turn, was negatively related to psycho- By collapsing together nodes that were structurally tributed across its members. In public health, net-logical measures of satisfaction and overall well- equivalent—i.e., those that had similar incoming work approaches have been important both inbeing. A similar study of 369 boys and 366 girls and outgoing ties—they could form a new network stopping the spread of infectious diseases and inbetween the ages of 13 and 19 in a Midwestern (a reduced model) in which the nodes consisted providing better health care and social of about 10,000 residents found that the of structural positions rather than individuals. Of all the applied fields, national security isadolescents’ behaviors were strongly influenced This idea mapped well with the anthropologists’ probably the area that has most embraced socialby the “cliques” to which they belonged (11). view of social structure as a network of roles network analysis. Crime-fighters, particularly thoseThe representation and analysis of community rather than individuals, and was broadly applica- fighting organized crime, have used a networknetwork structure remains at the forefront of net- ble to the analysis of roles in other settings, such perspective for many years, covering walls withwork research in the social sciences today, with as the structure of the U.S. economy (19). It was huge maps showing links between “persons ofgrowing interest in unraveling the structure of also noted that structurally equivalent individuals interest.” This network approach is often creditedcomputer-supported virtual communities that have faced similar social environments and therefore with contributing to the capture of Saddam Hussein.proliferated in recent years (12). In addition, terrorist groups are widely seen as By the 1960s, the network per- Chain networks rather than organizations, fueling researchspective was thriving in anthropol- Wheel Y Circle on how to disrupt functioning networks (26). Atogy. Influenced by the pioneering the same time, it is often asserted that it takes awork of Radcliffe Brown (13), there network to fight a network, sparking militarywere three main lines of inquiry. experiments with decentralized units.First, at the conceptual level, an-thropologists like S. F. Nadel began Social Network Theoryto see societies not as monolithic Perhaps the oldest criticism of social networkentities but rather as a “pattern or research is that the field lacks a (native) theo-network (or ‘system’) of relation- retical understanding—it is “merely descriptive”ships obtaining between actors in Centralized Decentralized or “just methodology.” On the contrary, there istheir capacity of playing roles rel- so much of it that one of the main purposes of thisative to one another” (14). Second, Fig. 2. Four network structures examined by Bavelas and article is to organize and simplify this burgeoningbuilding on the insights of the an- colleagues at MIT. Each node represents a person; each line body of theory. We will give brief summaries ofthropologist Levi-Strauss, scholars represents a potential channel for interpersonal communication. the salient points, using comparisons with thebegan to represent kinship systems The most central node in each network is colored red. network approach used in the physical sciencesas relational algebras that consisted (including biology).of a small set of generating relations (such as could be expected to develop similar responses, Types of ties. In the physical sciences, it is not“parent of” and “married to”) together with binary such as similar attitudes or behaviors (20). unusual to regard any dyadic phenomena as acomposition operations to construct derived re- Another key contribution was the influential network. In this usage, a network and a mathe-lations such as “in-law” and “cousin.” It was soon strength of weak ties (SWT) theory developed by matical graph are synonymous, and a commondiscovered that the kinship systems of such peoples Granovetter (21). Granovetter argued that strong set of techniques is used to analyze all instances,as the Arunda of Australia formed elegant math- ties tend to be “clumpy” in the sense that one’s from protein interactions to coauthorship to in-ematical structures that gave hope to the idea that close contacts tend to know each other. As a ternational trade. In contrast, social scientistsdeep lawlike regularities might underlie the ap- result, some of the information they pass along is typically distinguish among different kinds ofparent chaos of human social systems (15, 16). redundant—what a person hears from contact A dyadic links both analytically and theoretically. Third, a number of social anthropologists began is the same as what the person heard from B. In For example, the typology shown in Fig. 3 dividesto use network-based explanations to account for contrast, weak ties (e.g., mere acquaintances) can dyadic relations into four basic types—similarities,a range of outcomes. For example, a classic eth- easily be unconnected to the rest of one’s net- social relations, interactions, and flows. Much ofnographic study by Bott (17) examined 20 urban work, and therefore more likely to be sources of social network research can be seen as workingBritish families and attempted to explain the con- novel information. Twenty years later, this work out how these different kinds of ties affect eachsiderable variation in the way husbands and wives has developed into a general theory of social other.performed their family roles. In some families, capital—the idea that whom a person is connected The importance of structure. As in the studythere was a strict division of labor: Husband and to, and how these contacts are connected to each of isomers in chemistry, a fundamental axiom ofwife carried out distinct household tasks sepa- other, enable people to access resources that ulti- social network analysis is the concept that struc-rately and independently. In other families, the mately lead them to such things as better jobs and ture matters. For example, teams with the samehusband and wife shared many of the same tasks faster promotions (22). composition of member skills can perform veryand interacted as equals. Bott found that the By the 1980s, social network analysis had differently depending on the patterns of relation-degree of segregation in the role-relationship of become an established field within the social ships among the members. Similarly, at the levelhusband and wife varies directly with the con- sciences, with a professional organization (INSNA, of the individual node, a node’s outcomes and SCIENCE VOL 323 13 FEBRUARY 2009 893
  3. 3. REVIEW research area has been the pre- Similarities Social Relations Interactions Flows diction of similarity in time-to- Location Membership Attribute Kinship Other role Affective Cognitive e.g., e.g., adoption of an innovation for e.g., e.g., e.g., e.g., e.g., e.g., e.g., Sex with Information pairs of actors (31). Performance Same Same Same Mother of Friend of Likes Knows Talked to Beliefs refers to a node’s outcomes with spatial clubs gender and Sibling of Boss of Hates Knows Advice to Personnel respect to some good. For exam- temporal Same Same about events attitude Student of etc. Helped Resources ple, researchers have found that space Sees as etc. etc. Competitor of happy Harmed etc. firm centrality predicts the firm’s etc. etc. ability to innovate, as measured by number of patents secured (32), Fig. 3. A typology of ties studied in social network analysis. as well as to perform well finan- cially (33). Other research has future characteristics depend in part on its posi- formation of network ties and, more generally, to linked individual centrality with power and tion in the network structure. Whereas traditional predict a host of network properties, such as the influence (34). social research explained an individual’s outcomes clusteredness of networks or the distributions of Theoretical mechanisms. Perhaps the most or characteristics as a function of other character- node centrality. In the social sciences, most work common mechanism for explaining conse- istics of the same individual (e.g., income as a of this type has been conducted at the dyadic quences of social network variables is some form function of education and gender), social network level to examine such questions as: What is the of direct transmission from node to node. Whether researchers look to the individual’s social environ- basis of friendship ties? How do firms pick alli- this is a physical transfer, as in the case of mate- ment for explanations, whether through influence ance partners? A host of explanations have been rial resources such as money (35), or a mimetic processes (e.g., individuals adopting their friends’ proposed in different settings, but we find they (imitative) process, such as the contagion of ideas, occupational choices) or leveraging processes (e.g., can usefully be grouped into two basic categories: the underlying idea is that something flows along an individual can get certain things done because opportunity-based antecedents (the likelihood a network path from one node to the other. of the connections she has to powerful others). A that two nodes will come into contact) and benefit- The adaptation mechanism states that nodes key task of social network analysis has been to based antecedents (some kind of utility maximi- become homogeneous as a result of experiencing invent graph-theoretic properties that characterize zation or discomfort minimization that leads to tie and adapting to similar social environments. structures, positions, and dyadic properties (such formation). Much like explanations of convergent forms in as the cohesion or connectedness of the structure) Although there are many studies of network biology, if two nodes have ties to the same (or and the overall “shape” (i.e., distribution) of ties. antecedents, the primary focus of network research equivalent) others, they face the same environ- At the node level of analysis, the most widely in the social sciences has been on the consequences mental forces and are likely to adapt by becoming studied concept is centrality—a family of node- of networks. Perhaps the most fundamental axiom increasingly similar. For example, two highly level properties relating to the structural impor- in social network research is that a node’s position central nodes in an advice network could develop tance or prominence of a node in the network. in a network determines in part the opportunities similar distaste for the telephone and e-mail, For example, one type of centrality is Freeman’s and constraints that it encounters, and in this way because both receive so many requests for help betweenness, which captures the property of fre- plays an important role in a node’s outcomes. This through these media. Unlike the case of trans- quently lying along the shortest paths between is the network thinking behind the popular con- mission, the state of “distaste for communication pairs of nodes (27). This is often interpreted in cept of social capital, which in one formulation media” is not transmitted from one node to terms of the potential power that an actor might posits that the rate of return on an actor’s invest- another, but rather is similarly created in each wield due to the ability to slow down flows or to ment in their human capital (i.e., their knowledge, node because of their similar relations to others. distort what is passed along in such a way as to skills, and abilities) is determined by their social The binding mechanism is similar to the old serve the actor’s interests. For example, Padgett capital (i.e., their network location) (29). concept of covalent bonding in chemistry. The and Ansell (28) analyzed historical data on mar- Unlike the physical sciences, a multitude of idea is that social ties can bind nodes together in riages and financial transactions of the powerful node outcomes have been studied as conse- such a way as to construct a new entity whose Medici family in 15th-century Florence. The study properties can be different from suggested that the Medici’s rise to power was a those of its constituent elements. function of their position of high betweenness Binding is one of the mechanisms within the network, which allowed them to behind the popular notion of the broker business deals and serve as a crucial hub performance benefits of “structur- for communication and political decision-making. al holes” (Fig. 4). Given an ego- Research questions. In the physical sciences, network (the set of nodes with direct a key research goal has been formulating univer- ties to a focal node, called “ego,” sal characteristics of nonrandom networks, such Open Closed together with the set of ties among as the property of having a scale-free degree distri- members of the ego network), a bution. In the social sciences, however, researchers Fig. 4. Two illustrative ego networks. The one on the left structural hole is the absence of a tie have tended to emphasize variation in structure contains many structural holes; the one on the right contains few. among a pair of nodes in the ego across different groups or contexts, using these network (22). A well-established variations to explain differences in outcomes. For quences of social network variables. Broadly proposition in social network analysis is that example, Granovetter argued that when the city speaking, these outcomes fall into two main cat- egos with lots of structural holes are better per- of Boston sought to absorb two neighboring egories: homogeneity and performance. Node formers in certain competitive settings (19). The towns, the reason that one of the towns was able homogeneity refers to the similarity of actors lack of structural holes around a node means that to successfully resist was that its more diffuse with respect to behaviors or internal structures. the node’s contacts are “bound” together—they network structure was more conducive to collective For example, if the actors are firms, one area of can communicate and coordinate so as to act as action (21). research tries to predict which firms adopt the one, creating a formidable “other” to negotiate A research goal that the social and physical same organizational governance structures (30); with. This is the basic principle behind the ben- sciences have shared has been to explain the similarly, where the nodes are individuals, a key efits of worker’s unions and political alliances. In894 13 FEBRUARY 2009 VOL 323 SCIENCE
  4. 4. REVIEWcontrast, a node with many structural holes can value and not compared to expected values gen- This is one of many areas where we can each takeplay unconnected nodes against each other, divid- erated by a theoretical model such as Erdos-Renyi lessons from the and conquering. random graphs. For their part, social scientists References and Notes The exclusion mechanism refers to com- have reacted to this practice with considerable be- 1. M. Newman, A. Barabasi, D. J. Watts, Eds., The Structurepetitive situations in which one node, by forming musement. To them, baseline models like simple and Dynamics of Networks (Princeton Univ. Press,a relation with another, excludes a third node. To random graphs seem naïve in the extreme—like Princeton, NJ, 2006). 2. For a thorough history of the field, see the definitive workillustrate, consider a “chain” network (Fig. 5) in comparing the structure of a skyscraper to a random by Freeman (39).which nodes are allowed to make pairwise “deals” distribution of the same quantities of materials. 3. J. L. Moreno, Who Shall Survive? Nervous and Mentalwith those they are directly connected to. Node d More importantly, however, social and physi- Disease Publishing Company, Washington, DC, 1934).can make a deal with either node c or node e, but cal scientists tend to have different goals. In the 4. E. Durkheim, Suicide: A Study in Sociologynot both nodes. Thus, node d can exclude node c physical sciences, it has not been unusual for a (Free Press, New York, 1951). 5. R. D. Luce, A. Perry, Psychometrika 14, 95 (1949).by making a deal with node e. A set of exper- research paper to have as its goal to demonstrate 6. H. Leavitt, J. Abnorm. Soc. Psychol. 46, 38 (1951).iments (36) showed that nodes b and d have high that a series of networks have a certain property 7. Later experiments suggested that this result wasbargaining power, whereas nodes a, c, and e have (and that this property would be rare in random contingent on other factors. For example, severallow power. Of special interest is the situation of networks). For social scientists, the default expec- experiments showed that, as the complexity of puzzles increased, decentralized networks performed better (40).node c, which is more central than, and has as tation has been that different networks (and nodes 8. I. de S. Pool, M. Kochen, Soc. Networks 1, 1 (1978).many trading partners as, nodes b and d. How- within them) will have varying network proper- 9. S. Milgram, Psychol. Today 1, 60 (1967).ever, nodes b and d are stronger because each ties and that these variations account for differ- 10. C. S. Fischer, To Dwell Among Friends (Univ. of Chicagohave partners (nodes a and e) that are in weak ences in outcomes for the networks (or nodes). Press, Chicago, IL, 1948) 11. C. E. Hollingshead, Elmtown’s Youth (Wiley, London, 1949).positions (no alternative bargaining partners). Indeed, it is the relating of network differences to 12. B. Wellman et al., Annu. Rev. Sociol. 22, 213 (1996). outcomes that they see as constituting 13. R. Brown, Structure and Function in Primitive Society theoretical versus descriptive work. (Free Press, Glencoe, IL, 1952). Social scientists have also been 14. S. F. Nadel, The Theory of Social Structure a b c d e (Free Press, Glencoe, IL, 1957). more concerned than the physical 15. H. White, An Anatomy of Kinship: Mathematical Models scientists with the individual node, for Structures of Cumulated Roles (Prentice Hall,Fig. 5. A five-person exchange network. Nodes represent whether an individual or a collec- Engelwood, NJ, 1963).persons; lines represent exchange relations. 16. J. P. Boyd, J. Math. Psychol. 6, 139 (1969). tive such as a company, than with 17. E. Bott, Family and Social Network (Tavistock, London,Having only strong nodes to bargain with makes the network as a whole. This focus on node-level 1957).node c weak. In this way, a node’s power be- outcomes is probably driven to at least some 18. F. P. Lorrain, H. C. White, J. Math. Sociol. 1, 49 (1971).comes a function of the powers of all other nodes extent by the fact that traditional social science 19. R. S. Burt, Corporate Profits and Cooptation (Academicin the network, and results in a situation in which theories have focused largely on the individual. Press, NY, 1983). 20. R. S. Burt, Am. J. Sociol. 92, 1287 (1987).a node’s power can be affected by changes in the To compete against more established social sci- 21. M. S. Granovetter, Am. J. Sociol. 78, 1360 (1973).network far away from the node. An example of ence theories, network researchers have had to 22. R. S. Burt, Structural Holes: The Social Structure ofthe exclusion mechanism occurs in business-to- show that network theory can better explain the Competition (Harvard Univ. Press, Cambridge, MA. 1992).business supply chains. When a firm intentionally same kinds of outcomes that have been the tra- 23. R. Cross, A. Parker, The Hidden Power of Social Networkslocks up a supplier to an exclusive contract, ditional focus of the social sciences. (Harvard Business School Press, Boston, MA, 2004). 24. J. A. Levy, B. A. Pescosolido, Social Networks and Healthcompetitor firms are excluded from accessing Some physicists argue that direct observation (Elsevier, London, 2002).that supplier, leaving them vulnerable in the of who interacts with whom would be preferable 25. M. Sageman, Understanding Terror Networks (Univ. ofmarketplace. to asking respondents about their social contacts, Pennsylvania Press, Philadelphia, 2004). In quantum physics, the Heisenberg uncer- on the grounds that survey data are prone to error. 26. S. P. Borgatti, in Dynamic Social Network Modeling and Analysis: Workshop Summary and Papers, R. Breiger,tainty principle describes the effects of an observer Social scientists agree that survey data contain K. Carley, P. Pattison, Eds. (National Academy of Scienceson the system being measured. A foreseeable chal- error, but do not regard an error-free measurement Press, Washington, DC, 2003), p. 241.lenge for network research in the social sciences is of who interacts with whom to be a substitute for, 27. L. C. Freeman, Sociometry 40, 35 (1977).that its theories can diffuse through a population, say, who trusts whom, as these are qualitatively 28. J. F. Padgett, C. K. Ansell, Am. J. Sociol. 98, 1259 (1993). 29. R. S. Burt, Brokerage and Closure (Oxford Univ. Press,influencing the way people see themselves and different ties that can have different outcomes. In New York, 2005).how they act, a phenomenon that Giddens de- addition, social scientists would note that even 30. G. F. Davis, H. R. Greve, Am. J. Sociol. 103, 1 (1997).scribed as the double-hermeneutic (37). For exam- when objective measures are available, it is often 31. T. W. Valente, Soc. Networks 18, 69 (1996).ple, there has been an explosion in the popularity of more useful for predicting behavior to measure a 32. W. Powell, K. Koput, L. Smith-Doerr, Adm. Sci. Q. 41, 116 (1996).social networking sites, such as Facebook and person’s perception of their world than to measure 33. A. V. Shipilov, S. X. Li, Adm. Sci. Q. 53, 73 (2008).Linkedin, which make one’s connections highly their actual world. Furthermore, the varying abil- 34. D. J. Brass, Adm. Sci. Q. 29, 518 (1984).visible and salient. Many of these sites offer users ity of social actors to correctly perceive the net- 35. N. Lin, Annu. Rev. Sociol. 25, 467 (1999).detailed information about the structure and con- work around them is an interesting variable in 36. T. Yamagishi, M. R. Gilmore, K. S. Cook, Am. J. Sociol.tent of their social networks, as well as suggestions itself, with strong consequences for such outcomes 93, 833 (1988). 37. A. Giddens, The Constitution of Society (Univ. offor how to enhance their social networks. Will this as workplace performance (38). California Press, Berkeley and Los Angeles, 1984).enhanced awareness of social network theories It is apparent that the physical and social 38. D. Krackhardt, M. Kilduff, J. Pers. Soc. Psychol. 76, 770 (1999).alter the way in which people create, maintain, and sciences are most comfortable at different points 39. L. C. Freeman, The Development of Social Networkleverage their social networks? along the (related) continua of universalism to Analysis: A Study in the Sociology of Science (Empirical Press, Vancouver, 2004). particularism, and simplicity to complexity. From 40. M. E. Shaw, in Advances in Experimental SocialFinal Observations a social scientist’s point of view, network research Psychology, L. Berkowitz, Ed. (Academic Press, New York,A curious thing about relations among physical in the physical sciences can seem alarmingly sim- 1964), vol. 1, p. 111.and social scientists who study networks is that plistic and coarse-grained. And, no doubt, from 41. We thank A. Caster, R. Chase, L. Freeman, and B. Wellman for their help in improving this manuscript. This work was fundedeach camp tends to see the other as merely de- a physical scientist’s point of view, network re- in part by grant HDTRA1-08-1-0002-P00002 from thescriptive. To a physical scientist, network research search in the social sciences must appear oddly Defense Threat Reduction Agency and by the Gatton Collegein the social sciences is descriptive because mea- mired in the minute and the particular, using tiny of Business and Economics at the University of Kentucky.sures of network properties are often taken at face data sets and treating every context as different. 10.1126/science.1165821 SCIENCE VOL 323 13 FEBRUARY 2009 895
  5. 5. CORRECTIONS & CLARIFICATIONSERRATUM Post date 24 April 2009Reviews: “Network analysis in the social sciences” by S. P. Borgatti et al. (13 February, p. 892).On page 892, the final sentence in the legend for Fig. 1 was missing. The sentence should read:“Dashed lines represent mutual repulsion.” SCIENCE ERRATUM POST DATE 24 APRIL 2009 1